Icy ConditioningAverage

IcyConditioningAverage is the IcyHider clone of ConditioningAverage, and it's one of those small nodes you discover mid-project and immediately wish you'd found sooner: a fader between two conditionings. If you've ever wanted "prompt A but 30% closer to prompt B," this is the tool.

What it does

Two conditioning inputs and one dial:

• conditioning_to - the conditioning you're moving toward.
• conditioning_from - the one you're moving away from.
• conditioning_to_strength - 0.0 to 1.0, default 1.0. At 1.0 you get 100% "to" (a passthrough). At 0.0 you get 100% "from". In between, the node blends the two embedding sets.

Output is a single CONDITIONING for the sampler. Mechanically it's a weighted average of the two conditioning embeddings - the same idea as mixing LoRA weights or prompt interpolation, but at the conditioning level, before sampling, which makes it cheap and easily animated.

Where it's genuinely useful

• Dialing a style/prompt mix. Two prompts, one slider, find the balance. Want the mood of "noir, rain, neon" but keep "sunny park" as a safety rail? Average them at 0.7 and see.
• Prompt interpolation. Bump the strength over a sequence of fixed-seed generations and you get a smooth prompt morph - a classic animation trick.
• Softening a hard prompt. A prompt that's too specific can box the model in; averaging it against a looser one lets it breathe.

Where people get burned: it's easy to confuse the direction - "strength 0.8" reads like "mostly from" when it means "mostly to." Also, averaging only blends embeddings, so it won't resolve contradictions; the result is just a middle-ground prompt interpretation, which is exactly what you asked for and not a fix for prompt wars.

The "Icy" caveat

Outputs CONDITIONING - no preview for IcyHider to hide, so the wrapper is cosmetic packaging (the pack clones every core node at startup; this is one of the hundreds it relabels). The underlying ConditioningAverage is what actually runs. Use whichever name your graph template references.

Install

ComfyUI Manager → search "IcyHider" → install → restart, or:

cd ComfyUI/custom_nodes
git clone https://github.com/icekiub-ai/ComfyUI-IcyHider.git

Restart. No Python dependencies, no model downloads. The node lives in IcyHider Comfy Core. If the average seems to "do nothing," check which conditioning is wired to to vs from first - that's the whole node in one dial.